A general-purpose PyTorch codebase for 3D object detection with state-of-the-art model implementations and multi-dataset support.
Det3D is a general-purpose 3D object detection codebase built in PyTorch that provides implementations of state-of-the-art algorithms like PointPillars, SECOND, and VoxelNet. It solves the problem of fragmented research code by offering a unified toolbox with support for multiple datasets, enabling faster experimentation and benchmarking in 3D perception tasks.
Researchers and developers working on 3D perception for autonomous driving, robotics, or computer vision who need reproducible, high-performance detection models and a flexible codebase for experimentation.
Developers choose Det3D because it is the first comprehensive toolbox of its kind, offering out-of-the-box state-of-the-art models, multi-dataset support, and features like distributed training—all designed to accelerate research and development in 3D object detection.
World's first general purpose 3D object detection codebse.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Achieves top results on benchmarks like nuScenes and KITTI, with CBGS scoring 61.3 NDS and SECOND showing high AP on KITTI, as documented in the model zoo with provided checkpoints.
Supports KITTI, nuScenes, and Lyft datasets out-of-the-box, enabling diverse training and evaluation scenarios without extensive customization.
Includes distributed training with DDP and SyncBN, plus features like multi-task learning and rotated RoI align, facilitating robust experimentation for research.
Offers pre-trained models and configurations that ensure easy reproduction of published results, accelerating research cycles with clear benchmarks.
The TODO list acknowledges missing implementations like PointRCNN and PIXOR, and Waymo support is not fully integrated, limiting the range of available algorithms.
Installation and quick start require referring to separate markdown files (INSTALLATION.md, GETTING_STARTED.md), and the codebase assumes familiarity with advanced PyTorch and 3D detection concepts.
As a derivative project, it may have slower updates and less community support compared to broader frameworks like mmdetection3d, impacting long-term maintenance and feature additions.